ADSL Technology Explained, Part 1: The Physical Layer

Louis Litwin, Michael Pugel, Rob Rhodes, and John Richardson

Mar 01, 2001 (11:35 AM)
URL: http://www.commsdesign.com/showArticle.jhtml?articleID=16502655

For residential and commercial users with an ongoing need for broadband data access, but who do not send out correspondingly large data streams, asymmetric digital
subscriber line (ADSL) services work well. This service is so named because the data rate sent to the user (downstream) is much greater than the data rate sent from the user
(upstream). This asymmetric model is based on typical Internet usage patterns.

For example, a user sends a Web page request (small amount of upstream data) and receives the HTML for the Web page with graphics and sound (large amount of
downstream data). Various services, such as cable modems, satellite services, and DSL exist to provide such access.

An ADSL system uses existing telephone wire to allow bidirectional data communications between a user and the telephone company's central office (CO). Some other
popular services, such as an ISDN line or a standard dial-up modem, also use the phone lines to communicate. However, those services prevent the simultaneous operation of
standard analog phone service on the same phone line. An important advantage of ADSL is that it allows the plain old telephone system (POTS) signal to co-exist with the
ADSL data signal.

We begin our tour of the ADSL system architecture with a look at the physical layer (PHY). Topics covered at the PHY will include ADSL's multicarrier modulation technique,
common impairments, and phone-line characteristics.

Spectrum allocation

The ADSL PHY was designed so that it could peacefully co-exist with the standard POTS spectrum. The two services can co-exist because the ADSL spectrum only uses the
frequencies above POTS. The POTS spectrum goes from near DC to approximately 4 kHz. A frequency guard band is placed between the POTS spectrum and the ADSL
spectrum to help avoid interference. The ADSL spectrum starts above the POTS band and extends up to approximately 1.1 MHz. The lower part of the ADSL spectrum is for
upstream transmission (from the customer to the CO) and the upper part of the spectrum is fordownstream transmission. There are actually two different ways that the
upstream and downstream spectra can be arranged (see Figure 1).

In a frequency division multiplexed (FDM) system, the upstream and downstream spectra use separate frequency ranges. They can vary for different implementations, but
typically the upstream band is from 25 to 200 kHz and the downstream band is from 200 kHz to 1.1 MHz. Other divisions are also permitted within the ADSL standard. This
system is free from the occurrence of a type of interference called self-crosstalk. One drawback, however, is that the downstream bandwidth is reduced in comparison to an
echo-cancelled system.

An echo-cancelled system allows the downstream band to overlap with the upstream band. The upstream band still uses the frequencies from 25 to 200kHz, but the
downstream band can now extend over the upstream band. The main advantage of this system is that it significantly extends the available downstream bandwidth. However, it
does require echo-canceling circuitry due to the full-duplex transmission. In addition, the presence of self-crosstalk causes additional interference.

The DMT approach

The PHY of ADSL uses a multicarrier modulation technique known as discrete multitone (DMT). A DMT system transmits data on multiple subcarriers in a manner very similar
PDFmyURL.com
The PHY of ADSL uses a multicarrier modulation technique known as discrete multitone (DMT). A DMT system transmits data on multiple subcarriers in a manner very similar
to the orthogonal FDM (OFDM) technique that is used in many wireless applications. A DMT modulator takes in N data symbols in parallel and transmits the symbols on N
subcarriers. The data rate on each subcarrier is 1/ N the original data rate.

Reducing the data rate results in a DMT symbol period that is N times as long as the original symbol period. Increasing the symbol period can make the symbol longer than
the time span of the channel. This situation can make it easier to combat the effects of intersymbol interference.

The DMT signal is formed by using an Inverse Fast Fourier Trans-form (IFFT) to generate orthogonal subcarriers at the transmitter. The data symbols at the transmitter are
treated as being in the frequency domain and act as complex weights for the basis functions (orthogonal sinusoids at different frequencies) of the IFFT. The IFFT then converts
the data symbols into a time-domain "sum of sinusoids" signal.

The block of IFFT output samples is known as a DMT symbol. This time-domain signal is transmitted across the channel, and an FFT is used at the receiver to bring the
signal back into the frequency domain. A block diagram of a typical ADSL transmitter/receiver pair is shown in Figure 2.

A 2N-point IFFT is used to generate the DMT symbol, and the N negative-frequency IFFT bins are the complex conjugate of the N positive-frequency bins. This symmetric
spectrum results in a real time-domain signal. The DMT signal is centered at DC with the subcarriers around DC zeroed out (not used) to create a hole in the DMT spectrum in
order to make room for the POTS spectrum. DMT is thus a true baseband system.

DMT supports inclusion of a cyclic prefix. A cyclic prefix is a block of samples with a length, LP, that is a replica of the last L P samples of the DMT symbol. The prefix is then
transmitted first, followed by the 2N samples of the DMT symbol. The length LP is chosen such that it will be longer than the length of the channel response. The cyclic prefix
contains redundant information. However, the DMT receiver exploits the presence of the prefix in order to mitigate the effects of the channel. The use of the cyclic prefix will be
described in further detail in the Impairments section.

The dynamic bit allocation technique allows DMT to make efficient use of the available channel capacity. This technique enables the system to vary the number of bits per
symbol for each subcarrier based on the subcarrier's signal-to-noise ratio (SNR). Subcarriers with a low SNR transmit binary phase-shift keying (BPSK) or quadrature PSK
(QPSK) because they are robust modulation formats. If the subcarrier's SNR is very low, that subcarrier will not be used to transmit data at all. Subcarriers with a higher SNR
transmit higher-order quadrature amplitude modulation (QAM) in order to achieve an increased throughput.

Impairments

There are several significant types of impairments encountered in an ADSL system: additive white Gaussian noise (AWGN), crosstalk, impulse noise, bridged taps, and radio
noise.

AWGN is the thermal noise that is common to all communication systems. In a digital system such as ADSL, AWGN can cause symbol errors to occur at the receiver when
noise pushes the received sample beyond a decision boundary. Like many other digital communication systems, ADSL employs error-control coding to help mitigate the effect
of AWGN. Coding adds redundancy to the transmitted signal and exploits the redundancy at the receiver to detect and correct errors.

ADSL uses three layers of coding. The innermost code in the PHY is a convolutional code. These codes get their name because the encoding process can be viewed as the
convolution of the message with the code's impulse response. The Viterbi algorithm is used at the receiver to decode the received sequence.

Convolutional codes

Convolutional codes are good at correcting random errors. However, the nature of the decoding algorithm is such that the decoder can cause burst errors to occur if errors are
made during the decoding process.

A Reed-Solomon block code is used on top of the convolutional code. Reed-Solomon codes are powerful codes that are good at detecting and correcting burst errors, such as
those generated by the Viterbi decoder. The ADSL specification allows Reed-Solomon code-word lengths of up to 255 bytes with the addition of up to 16 parity bytes for each
code word. The outermost code is a cyclic redundancy check (CRC) code. The CRC can detect errors, but it cannot correct them. The CRC code is used as a top-level error-

PDFmyURL.com
detection mechanism in order to detect any errors that remain after Viterbi and Reed-Solomon decoding.

Because bundled telephone cable contains many wires for many different users, crosstalk is a common impairment. These wires radiate electromagnetically and can induce
currents in other wires in the cable. This interference effect is known as crosstalk. There are two basic types of crosstalk and they both appear at the receiver as additive
noise.

Near-end crosstalk (NEXT) occurs when a transmitter interferes with a receiver located on the same end of the cable. Far-end crosstalk (FEXT) occurs when the transmitter
interferes with a receiver on the opposite end of the cable. The effect of NEXT is more severe than FEXT since the FEXT interference travels the entire length of the cable and
is attenuated by the time it reaches the receiver.

Crosstalk can be further subdivided into self-crosstalk and foreign crosstalk. Self-crosstalk is interference from another ADSL system using the same spectrum allocation.
Foreign crosstalk is interference from an ADSL system using a different spectrum allocation or from a completely different type of system (such as ISDN). One way to reduce
the effects of crosstalk is with spectrum allocation.

In echo-cancelled ADSL, the upstream and downstream channels overlap (see Figure 1). Since the same frequency band is being used for transmission and reception, the
system will suffer from self- and foreign crosstalk. However, in FDM ADSL the upstream and downstream channels use separate frequency bands. This system will not suffer
from self-crosstalk, although foreign crosstalk will still be present.

Interference that is short in duration but of a large magnitude is known as impulse noise. Impulse noise can be caused by lightning or by a motor turning on and creating a
power surge. ADSL systems use a combination of interleaving and coding to correct the errors caused by impulse noise. The interleaving process rearranges data so that
those samples that were located contiguously in time are spaced far apart. Impulse noise can cause a burst of errors that is hard for the decoders to correct. The use of
interleaving combined with coding spreads out these errors in time to improve decoding performance.

An additional impairment

Bridged taps are an additional impairment type found in ADSL systems. A bridged tap is a section of wire connected to the loop at one end and unterminated at the other end.
Examples of bridged taps are unterminated wires that are laid out in areas where housing is still being built. When a transmitted signal arrives at a bridged tap, the signal
divides. Part of the energy continues to the receiver and the rest of the energy reflects off of the unterminated end. This reflection causes delayed versions of the signal to
arrive at the receiver, and these reflections distort the received signal. The effect is very similar to the interference that occurs with a multipath channel in a wireless
communication system.

Bridged taps cause two problems. The first is intersymbol interference where a received DMT symbol is distorted due to delayed versions of the previous DMT symbol. The
effect of intersymbol interference is removed by discarding the cyclic prefix at the receiver. The cyclic prefix contains redundant information, and so it is not needed at the
receiver. The length of the prefix is chosen such that the delayed versions of the previous symbol only distort the cyclic prefix and not the actual data part of the DMT symbol.
The ability to remove the effect of intersymbol interference by discarding the cyclic prefix is one of the advantages of having a long symbol period.

The second problem is the intrasymbol interference that is caused when delayed versions of a DMT symbol cause the symbol to interfere with itself. The effect of intrasymbol
interference in the frequency domain is a shaping of the received-signal spectrum. The received spectrum is essentially a multiplication of the transmitted spectrum and the
channel's frequency response. The distortion due to intrasymbol interference is removed by using a frequency-domain equalizer.

And finally, radio noise is interference due to a wireless source. The copper phone lines act as antennae and pick up this interference. The most common source of radio noise
comes from AM radio since its spectrum overlaps with the ADSL spectrum. Coding can help correct the errors caused by radio noise, and adaptive RF cancellation filters can
also be used. The previously mentioned technique of dynamic bit allocation can be used to turn off subcarriers near frequencies of interference.

ADSL in the physical plant

ADSL operates within the existing POTS plant structure. The phone company has been in existence for nearly 100 years and despite modernization, the structure today
remains much as it did years ago. There are more than 700 million phone lines in the world with two-thirds in the US alone. Figure 3 illustrates the basic structure that is
PDFmyURL.com
utilized in many US phone-service deployments.

Mainline routes shown on the left side of Figure 3 are connected by fiber between toll offices (TOs). These TOs provide the interface between local exchanges and long-haul
fiber runs for long-distance service. The TOs also connect COs that control local exchange service. The CO houses the main switching equipment for the home and also
serves as the location for the head-end DSL equipment.

Multiple interconnects are often made between several COs and TOs for operational reliability. From the CO, the network is further subdivided into customer service areas
(CSAs) that are serviced by remote terminals (RTs). This region represents the last several miles before the home, such as a large neighborhood or small city.

The CSA is then further subdivided into a distribution area (DA) that is initiated at a feeder distribution interface (FDI) serving up to 500 phone lines. The FDI represents the last
point where the phone lines are still bundled together. The phone wires take separate routes to each home and typically cover the last mile of service routing.

Standard copper twisted pair is the primary medium for routing between the CO and RT, and then on to the home residence. Fiber deployment through these regions is
technically possible but economically unjustifiable, especially for points past the FDI. Fiber routing will eventually be used for routes between the CO and RT. This upgrade will
support improvements in ADSL service as well.

The ability to supply phone service over a very long distance of wire, often called reach, is limited by the ability of the phone switching equipment at the CO to function
correctly. The network switch can operate with a maximum DC load resistance of approximately 1,500 . Most of this resistance is found in the copper lines running between
the CO and home.

The distance at which this resistance is reached is known as the revised resistance design (RRD) distance. Voice quality over the phone often suffers prior to reaching the
RRD distance, which can extend to more than 3 miles. The phone company employs tricks to improve quality, including thicker wire and load coils. ADSL performance is also
a function of the distance from the CO and service is often extremely limited or impossible on these long reaches.

Plain old copper

The copper pair wire that runs between the CO and the home residence has the greatest impact on ADSL system performance. The wire provides simultaneous signal routing
of bidirectional voice communications, which will occupy the spectrum up to approximately 4 kHz. Above this frequency, the ADSL signals are inserted. The wire is best
modeled as a lossy transmission line. Most plants employ one of four copper wire gauges ranging from 19 to 26 gauge.

Thinner wire (higher gauge number) is typically used close to the CO to allow smaller bundles to be formed. Thicker wire is used close to the residence to extend the reach
from the CO to home while keeping within the resistance limits required for proper voice switching.

The transmission-line model contains a shunt capacitance in parallel with a shunt resistance (due to dielectric losses) along with a series resistance and series inductance.
The cable specifications dictate that the capacitance (per mile) is constant for all wire gauges. As a result, the nominal line impedance is typically around 120 . The series
resistance changes by approximately 20% over the possible wire gauges and increases logarithmically with frequency. Typical loss through the wire is 3 to 6 dB per mile in the
voice band, depending on wire gauge, and also increases logarithmically with frequency.

Most system impairments are created by, or enter the system through, the copper wire. The wire, through its lossy nature, accounts for degradation in SNR. Reflections are
created in the system due to unterminated lines such as bridged taps and wire splices. These reflections create self-interference. Reflections cause energy cancellation at
certain frequencies as discussed in the Impairments section. The wire is typically unshielded from the FDI (point of fanout) to the home location and is susceptible to ingress
(pickup of external signals in the frequency range of interest).

Part two of this article will complete the discussion of PHY equipment and move on to discuss the link layer. Other elements of the PHY plant will be introduced with a focus
on the equipment at the home. Link layer topics include ADSL modem initialization and packet framing. ATM and IP will be covered in the network-layer section. Finally, the
application layer will address some new services such as voice over DSL (VoDSL) and video streaming.

---
PDFmyURL.com
Rob Rhodes is manager of the Communications Design group at Thomson Multimedia (Indianapolis, IN). The group is responsible for the development of the PHY
interfaces for satellite, cable, and DTV communications. He can be reached at rhodesr@tce.com.

Mike Pugel is a principal member of the technical staff at Thomson Multimedia (Indianapolis, IN). He is currently work ing on advanced communication receiver concepts.
He can be reached at pugelm@tce.com.

Louis Litwin is a member of the technical staff at Thomson Multimedia (Princeton, NJ). His focus is the development of wireless communication devices used for digital
home network ing and mobility applications. He can be reached at litwinl@tce.com.

John Richardson is a member of the technical staff at Thomson Multimedia (Princeton, NJ). He is primarily involved in the areas of digital home network ing and ADSL
system development. He can be reached at richardsonj@tce.com.

C opyright © 2003 C MP Me dia , LLC | P riva cy Sta te m e nt

PDFmyURL.com